Interactive Virtual Angioscopy Enrico Gobbetti Piero Pili Antonio Zorcolo Massimiliano Tuveri, M.D. CRS4 Center for Advanced Studies, Research and Development in Sardinia Cagliari, Italy Abstract plaque, and its relationship with the rest of the vessel. Technolog- ical advances and improvements in the characterization of the dis- Virtual angioscopy is a non invasive medical procedure for explor- eased vessel as well as in the understanding of the precise spatial ing parts of the human vascular system. We have developed an in- location of the lesion to be stented are susceptible to increase the teractive tool that takes as input data acquired with standard med- success rate of vascular procedures and the feasibility of approach- ical imaging modalities and regards it as a virtual environment to ing complex lesions, thus reducing the rate of restenosis. be interactively inspected. The system supports real-time naviga- Recent advances in non-invasive data acquisition technology, tion with stereoscopic direct volume rendering and dynamic endo- e.g. spiral computerized tomography (CT) and fast magnetic res- scopic camera control, interactive tissue classi®cation, and interac- onance (MR) imaging, are now providing physicians with large tive point picking for morphological feature measurement. In this amounts of quantitative information on a patient's internal struc- paper, we provide an overview of the system, discuss the techniques tures, making it possible to carefully study parts of the arterial tree used in our prototype, and present experimental results on human before intervention. The ability to analyze scalar volume datasets data sets. The accompanying video-tape illustrates our approach acquired by the use of non invasive techniques is thus gaining im- with interactive sequences showing the examination of a human portance in the diagnostic and therapy planning ®elds, and physi- carotid artery. cians are now faced with the need to interpret data produced by CT Keywords: Virtual angioscopy, endoscopy, interactive rendering, and MR scanners. volume rendering, virtual environment Data analysis is currently done mostly by analyzing static images of 2D volume slices. Physicians are thus forced to mentally com- bine multiple views for constructing a mental model of 3D shapes, 1 Motivation adding further complexity to what is often an already dif®cult task. The creation of a virtual environment for the direct analysis in three Atherosclerosis is a generalized process in which morbidity and dimensions of volumetric data, i.e. a virtual angioscopy system, mortality usually result from localized plaque deposition rather than overcomes these problems [12, 7]. diffuse diseases. The locations of the depositions in susceptible In virtual angioscopy, and more generally virtual endoscopy, a zones of the arterial tree are relatively constant and predictable, three-dimensional dataset of a body part derived from diagnostic which suggests that vessel geometry and ¯uid localizing factors imaging data is displayed on a computer and made available for play an important role in the onset of the disease. Differing lumen 3D navigation and analysis. Research virtual endoscopy prototypes diameters, curvatures, branching, and angles produce local distur- are being developed and have proven useful for a variety of clin- bance in the primary blood ¯ow ®eld. Branch points are known ical applications (e.g. bronchoscopy [23], cholonoscopy [10] and to be particularly vulnerable to plaque formation and are subject to minimally invasive surgery planning [1]). wide variation in hemodynamic conditions. Emerging approaches Because of the computational costs associated with image pro- to vascular disease treatment are percutaneous transluminal angio- duction, work in this ®eld has until recently concentrated on gen- plasty and/or intravascular stenting after angioplasty. The effective- erating precomputed ¯y-through animations [12, 19, 22], which ness of these techniques remains however limited by a 30% to 50% makes these tools rather cumbersome to be used for the kind of restenosis rate within the ®rst six months. One of the reasons of detailed clinical examinations that are needed to examine com- this high failure rate is the dif®culty to characterize the arterial le- plex vessel lesions. The continued improvement and prolifera- sion, the precise three-dimensional geometry of the atherosclerotic tion of graphics hardware for workstations and personal computers promises to bring increasing prominence to a newer style of vir- [email protected] tual endoscopy system, which support real-time exploration and are [email protected] therefore better suited for virtual angioscopy applications. How- [email protected] ever, current interactive virtual endoscopy systems are mostly based [email protected] on surface rendering techniques, where volumetric data is automat- CRS4, Via N. Sauro 10, 09123 Cagliari, Italy, http://www.crs4.it ically or manually segmented during a preprocessing step to extract boundary surfaces that are then rendered during navigation [23, 10]. These techniques have proven to be often not accurate enough for clinical examinations [29, 16, 7]. In particular, since when using surface rendering only the interior surfaces of a cavity are visi- ble during navigation, diagnostic capabilities are severely limited [17, 18, 22]. Our aim is to merge the concepts of interactive direct volume rendering and virtual endoscopy in a tool for non-invasive visual- ization and analysis of diseased vessels. In this paper, we provide an overview of the system, discuss the techniques used in our pro- totype, and present experimental results on explanted human data sets. The accompanying video-tape illustrates our approach with their distance to the viewer, a good strategy is to space the interactive sequences showing the examination of human carotid slices proportionally to the inverse of that distance. With this artery CT data. approach spatial resolution is increased close to the viewer, making close-up inspection more precise while reducing the details in the background, which is appropriate for an inter- 2 System Overview active tool. As emission/absorption varies non linearly with thickness [13], special care has to be taken to correctly com- Our interactive virtual angioscopy system takes as input data ac- pute the contribution of each slice. In our renderer, we asso- quired with standard medical imaging modalities and regards it as ciate to each slice a thickness which depends on slice spac- a virtual environment to be interactively inspected. As the system ing. Texture lookup tables, a Silicon Graphics extension to employs the DICOM standard for input dataset coding, it is possible OpenGL, are then used to de®ne opacity values that depend to have it connected directly to standard medical data acquisition on slice thickness. Serra et al. also use variable slice spacing, devices. but without opacity correction, introducing notable artifacts Our laboratory is connected via LAN through a 2 Mbit/s ded- [21]; icated line to a CTN Image Server [28] that can acquire datasets from a Picker CT Scanner equipped with a DICOM Storage Class zoom rendering: pixel ®ll-rate is the major limiting factor Provider located in the Radiology Department of the S. Michele when using a texturing approach to volume visualization. In hospital [8]. New acquisitions are made thus available for exami- zoom rendering, an appropriately down-scaled image is ren- nation in a few minutes. When the dataset is loaded on the graphical dered in the back buffer and then enlarged and copied to the workstation, the system presents a perspective view of the dataset front buffer [14]. This way, delays associated with buffer and users may interactively take morphological measures, move the swap synchronization are avoided, and the number of pix- viewpoint, and modify the classi®cation parameters to obtain the els ®lled during volume rendering is reduced. In our imple- best visual representation of the particular anatomy under examina- mentation, the copy and zoom operations are implemented by tion. copying the reduced size image in texture memory and then When building a virtual angioscopy system, the most important rendering a textured polygon in the front buffer. This way, design factors to be taken into account are rendering and interaction sophisticated texture interpolation algorithms can be used to techniques. Our choices are discussed in the following sections. reduce the artifacts caused by magni®cation. Zoom rendering is particularly useful for endoscopic volume rendering appli- 2.1 Rendering Techniques cations, because the pixel resolution is much larger than the resolution of the data that is displayed in the window. Zoom Direct volume visualization is a rendering technique which does not rendering is only active during navigation, still images being require the construction of intermediate geometric representations recomputed at maximum resolution. and has the advantage of supporting the representation in a single image of the full dataset using semi-transparent mapping [20, 26, The ability to render perspective views of a volume at interac- 6]. tive speeds dramatically improves depth perception, thanks to mo- With this technique, images are produced by integrating along tion parallax effects, and makes it possible to use direct interaction selected projectors the value of a continuous absorption-emission techniques to naturally perform complex 3D tasks, such as ®nd-